Every consulting firm in the country added "AI-powered" to their pitch deck in the last eighteen months. Most of them are lying. Not in the legal sense — they have all used a chatbot once or twice — but in the sense that matters to a candidate trying to figure out which firm to write a check to.
The lie is a shape, not a sentence. It is a website with new graphics, a paragraph about "innovation," and a sales call that pivots back to the same five services the firm has sold for fifteen years. The firm bought a license. The firm did not build a doctrine. There is a difference between those two things. The difference is the entire conversation.
This briefing gives you a way to find out which one you are sitting across from. We call it The Pretender Test. It is four questions. It can be run in twenty minutes on a discovery call. The answers cannot be faked by anyone who has not actually done the work.
Question one. Can the firm show you a tool it built — not a tool it bought?
A firm that has done the work has at least one piece of software it built for the campaigns it runs. Not a workflow in someone else's product. Not a custom GPT. A tool. A platform. A system with a name, a deployment history, and a track record of failure modes the firm has solved in public.
If the firm cannot show you the tool, the firm does not have one. "We use AI throughout our process" is the answer of a firm that has not been past the gate. Dark Horse Political built ARMR before any client asked for it because the doctrine required it. The tool exists because the work demanded it. That is the only reason any tool worth using gets built.
Question two. Can the firm tell you what AI cannot do — and refuse to use it where it cannot?
This is the question that filters more pretenders than any other.
A firm using AI seriously knows where the model fails. They know what an AI will hallucinate, what it will sycophant, what it will drift on, and what categories of decision they will not let it near. They will tell you, unprompted, that they do not let AI write the candidate's positioning, score the candidate's stamina under fire, or pick the moment to throw the punch in a debate exchange.
A firm that pretends to use AI has no answer to this question. They have never let one near a live campaign and they do not know what the failure modes look like. They will say something like "we use it carefully" and stop talking. That is the tell.
The doctrine we operate by has a name for what happens when AI is allowed to make decisions it is not qualified to make. We call it Off-Doctrine Drift. Every firm needs to be able to talk about its version of that boundary. If they cannot describe the boundary, the boundary does not exist.
Question three. Can the firm walk you through one campaign decision they made differently because of the AI work — and what the outcome was?
This is the verification question. AI is not magic. It is a force multiplier. If it is being used seriously, it changes the campaign's decisions.
There is a moment in a serious AI workflow when a model surfaces a pattern in the comments under the opponent's town hall video that nobody on the staff saw. There is a debate prep session where a synthetic adversary throws a question the candidate had not prepared for and the candidate practices the response forty times before the real moderator asks it. There is a precinct call where the field data and the modeled propensity converge on a target the campaign would otherwise have missed.
If the firm cannot tell you what decision the AI changed, the firm did not use it. They watched it run. There is a difference.
Question four. Will the firm let you see how the tool actually runs?
A firm that has built something will show you. A firm that has not built anything will tell you the demo is proprietary. Both are technically true statements. Only one of them belongs to a firm you should hire.
The most useful version of this question is the smallest one: ask the firm to log into the tool on the call with you. Watch them navigate. Watch the friction. A tool that gets used has friction the operator has internalized — they click past errors without thinking, they have a keyboard shortcut for the thing they do every day, they make a small joke about a feature that broke once and got fixed. A tool the firm pretends to use has no such pattern. The operator clicks like a tourist.
What the test is actually for.
The Pretender Test does not exist because AI is the question of this election cycle. AI is the wedge. The question of this cycle is which firms have actually been doing the work, and which firms are running the old plays with new graphics. The four questions above are diagnostic for that larger problem. They use the AI claim because the AI claim is the easiest one to test. A firm that fails on AI is a firm that will fail on something else when the cycle gets serious.
Most candidates discover this in March of an election year, after the firm has been retained and the campaign has been built around services that do not exist. We are publishing the test because we would rather see a candidate hire a firm that passes it — even if that firm is not us — than watch them hire one that fails. The category needs the floor raised.
The candidates who are paying attention know this already. The ones who are not will figure it out the same way every cycle's lost campaigns figure it out — late, expensively, and on someone else's clock.
Run the test before the wire goes out. The right firm will pass it without needing to know you ran it.